A Spanish Financial Market Stress Index (FMSI) - Leticia Estévez Cerqueira Mª Isabel Cambón Murcia

Page created by Frederick Newman
 
CONTINUE READING
A Spanish Financial Market Stress Index (FMSI) - Leticia Estévez Cerqueira Mª Isabel Cambón Murcia
A Spanish Financial Market
Stress Index (FMSI)
Leticia Estévez Cerqueira
Mª Isabel Cambón Murcia

Documentos de Trabajo
No 60
A Spanish Financial Market Stress Index (FMSI)

Leticia Estévez Cerqueira
Mª Isabel Cambón Murcia

Working paper
No. 60
April 2015
Mª Isabel Cambón Murcia and Leticia Estévez Cerqueira are both members of the Research, Statistics and
Publications Department, CNMV.

The opinions in this Working Paper are the sole responsibility of the authors and they do not necessarily
coincide with those of the CNMV.

The CNMV publishes this Working Paper Series to enhance research and contribute towards greater
knowledge of the stock markets and their regulation.

The CNMV distributes its reports and publications via the internet at www.cnmv.es

© CNMV. The contents of this publication may be reproduced, subject to attribution.

Mª Isabel Cambón Murcia y Leticia Estévez Cerqueira pertenecen al Departamento de Estudios, Estadísticas
y Publicaciones de la CNMV.

Las opiniones expresadas en este documento relejan exclusivamente el criterio de los autores y no deben ser
atribuidas a la Comisión Nacional del Mercado de Valores.

La Comisión Nacional del Mercado de Valores, al publicar esta serie, pretende facilitar la difusión de estudios
que contribuyan al mejor conocimiento de los mercados de valores y su regulación.

La Comisión Nacional del Mercado de Valores difunde la mayoría de sus publicaciones a través de la red
Internet en la dirección www.cnmv.es

© CNMV. Se autoriza la reproducción de los contenidos de esta publicación siempre que se mencione su
procedencia.

ISSN (edición electrónica): 2172-7147

Maqueta: Composiciones Rali, S.A.
Abstract

The relevance of systemic risk was highlighted by the economic and financial crisis
starting in mid-2007. Supervisors and regulators recognised the need to improve the
process of identification, management and mitigation of systemic risk. This paper
introduces a Spanish Financial Market Stress Indicator (FMSI), similar to the “Com-
posite Indicator of Systemic Stress” that Holló, Kremer and Lo Duca (2012) pro-
posed for the euro area as a whole. This indicator, which represents a real-time
measure of systemic risk, tries to quantify stress in the Spanish financial system and
describes the contribution of each financial market segment (bond market, equity
market, money market, financial intermediaries, forex markets and derivatives) to
the total stress in the system. The methodology takes into account time-varying cor-
relations between market segments. The study analyses the ability of the FMSI to
identify past periods of high financial stress and presents two econometric ap-
proaches with the aim of classifying observations into different stress regimes and
of determining if financial stress has a negative impact on the real economy.

Keywords: systemic risk, financial crisis, composite indicator, real economy

JEL Classification: G01, G10, G20, E44.

                                                 A Spanish Financial Market Stress Index (FMSI)   5
Table of contents

1     Introduction                                              11

2     Theoretical background and related literature             13

3     Statistical design of the Spanish FMSI                    17

3.1   Selection of markets and variables                        17

3.2   Construction of market sub-indices                        20

3.3   Aggregation of sub-indices into the composite indicator   22

3.4   Backward extension                                        23

3.5   Robustness analysis                                       26

4     Evaluation of the Spanish FMSI                            31

4.1   Ability to identify stress events                         31

4.2   Regimes and thresholds                                    33

4.3   Evaluation of potential real effects                      38

5     Conclusions                                               43

Bibliography                                                    45

Annexes                                                         49
Index of figures

FIGURE 1    Spanish financial system structure and supervision scheme                    16

FIGURE 2    Cross-correlations between sub-indices                                       23

FIGURE 3    Financial Market Stress Indicator (FMSI)                                     24

FIGURE 4    Backward-extended proxy-FMSI                                                 25

FIGURE 5    FMSI versus hypothesis of perfect correlation                                25

FIGURE 6    Sub-indices aggregation by principal component analysis                      26

FIGURE 7    Market weighted FMSI                                                         27

FIGURE 8    Changes in the smoothing parameter                                           28

FIGURE 9    Backward-extended proxy-FMSI: recursive versus non-recursive                 29

FIGURE 10   Spanish FMSI and major financial stress episodes                             31

FIGURE 11   Histogram for the FMSI                                                       33

FIGURE 12   Fitted values and residuals for the MS(3)-DR(1) model                        35

FIGURE 13   Smoothed regime probabilities for the MS(3)-DR(1) model                      36

FIGURE 14   Spanish FMSI and regimes of stress from MS(3)-DR(1) model                    38

FIGURE 15   Spanish FMSI and industrial production                                       39

FIGURE 16   Spanish FMSI and regimes of financial stress                                 40

FIGURE 17   Impulse response functions (IRF) of industrial production growth to shocks
            in the FMSI from TVAR model                                                  41

FIGURE A1   Scatter plot of the Spanish FMSI (two months lagged) against industrial
            production (annual change)                                                   50

FIGURE A2   Impulse response functions (IRF) of exports growth to shocks in the FMSI
            from TVAR model                                                              50
Index of tables

TABLE 1    Testing different specifications of Markov-switching models for the Spanish FMSI   35

TABLE 2    Parameter estimates of the MS(3)-DR(1) model                                       37

TABLE 3    Transition matrix of the MS(3)-DR(1) model                                         37

TABLE 4    Testing the VAR versus TVAR and the threshold delay                                40

TABLE 5    Parameter estimates of TVAR (two thresholds, three regimes)                        42

TABLE A1   Summary of statistics                                                              49
1     Introduction

The global economic and financial crisis that many economies suffered after the col-
lapse of Lehman Brothers in 2008 highlighted the importance of systemic risk. Fol-
lowing the crisis, authorities and financial supervisors realized that the identifica-
tion of systemic risks deserved more attention. There was also a revision to the
definition of systemic risk published by international institutions (IMF, FBS, BIS
and IOSCO). One of the main lessons of this process was the recognition of the role
that both banking and securities regulators had to play in this area. There have been
many and various studies looking at some aspect of systemic risk in recent years. In
general, current research is related to one or more relevant factors when considering
systemic risk: size, interconnectedness, lack of substitutes and concentration, lack of
transparency, leverage, market participant behaviour, information asymmetry and
moral hazard.

There is a group of papers that, with the objective of measuring systemic risk, have
developed Financial Stress Indexes (FSI) or fragility indexes. Some of these are co-
incident measures (like thermometers) that try to capture the level of financial stress
in real time and others are forward-looking indicators. Other approaches have in
common the definition of systemic risk as an extreme loss on a portfolio of assets
related to financial intermediaries’ balance sheets. This definition of systemic risk
focuses on the financial health of intermediaries, rather than on monetary and cred-
it conditions. Finally, during the global financial and economic crisis, and especially
in the context of the European sovereign debt crisis, many studies focused on the
phenomenon of contagion.

This paper introduces a Spanish Financial Market Stress Indicator (FMSI), similar
to the “Composite Indicator of Systemic Stress” that Holló, Kremer and Lo Duca
(2012) proposed for the euro area as a whole. This kind of indicator, which can be
included in the group of Financial Stress Indicators (FSI), represents a coincident
measure of systemic risk and tries to quantify and summarize the stress in the Span-
ish financial system in a single statistic. As well as summarizing the statistical de-
sign of the indicator, we provide a threefold evaluation of the FMSI and propose
some applications in the context of the CNMV’s supervisory duties.

The remainder of the paper is structured as follows: section 2 summarizes the back-
ground and academic literature regarding systemic risk and explains the motivation
for this paper. Section 3 provides the details of the statistical design of the Spanish
FMSI, including the selection of markets and variables, the construction of the sub-
indices and their aggregation into the composite indicator. Section 4 evaluates the
indicator in terms of its ability to identify past episodes of stress in the Spanish fi-
nancial system. This section also presents the results of two econometric approach-
es related to the theory of switching regimes and to the potential impact of financial
stress on domestic output. Finally, section 5 lays out the main conclusions.

                                                  A Spanish Financial Market Stress Index (FMSI)   11
2      Theoretical background and related literature

Following the global financial crisis, which started by mid-2007, international au-
thorities and governments realized that financial stability analysis and the process
of identification of systemic risks should receive more attention. In their conclu-
sions it was clear that both banking and securities regulators had to play a role in
this area. In 2009, the International Monetary Fund (IMF), the Financial Stability
Board (FSB) and the Bank of International Settlements (BIS) set out an approach to
assessing the systemic importance of financial institutions, markets and instru-
ments. These institutions described systemic risk as:

    “[…] the risk of disruption to financial services that is (i) caused by an impair-
     ment of all or parts of the financial system and (ii) has the potential to have
     serious negative consequences for the real economy1.”

In 2010 the Board of the International Organization of Securities Commissions
(IOSCO) adopted two new principles (6 and 7) related to the process of monitoring,
mitigating and managing systemic risk and to the process of reviewing the perime-
ter of regulation. Moreover, in 2011, IOSCO published a definition of systemic risk
very close to that of IMF/FSB/BIS:

    “Systemic risk refers to the potential that an event, action, or series of events or
     actions will have a widespread adverse effect on the financial system and, in
     consequence, on the economy2”.

However, IOSCO elaborated on this definition, enumerating several factors which
potentially can increase systemic risk. They mentioned the design, distribution or
behaviour under stressed conditions of certain investment products, the activities or
failure of a regulated entity, a market disruption or an impairment of a market’s
integrity. From IOSCO’s perspective systemic risk can also take the form of a more
gradual erosion of market trust caused by inadequate investor protection standards,
lax enforcement, insufficient disclosure requirements, inadequate resolution re-
gimes or other factors.

The academic research community has pursued a plentiful variety of approaches in
the area of financial stability. In general, academic research has concentrated on
one or more relevant factors to consider when assessing systemic risk: size, inter-

1   “Guidance to Assess the Systemic Importance of Financial Institutions, Markets and Instruments: Initial
    Considerations”, International Monetary Fund, Bank for International Settlements and Financial Stability
    Board, October 2009.
2   “Mitigating Systemic Risk: A Role for Securities Regulators”, IOSCO, February 2011. http://www.iosco.org/
    library/pubdocs/pdf/IOSCOPD347.pdf

                                                              A Spanish Financial Market Stress Index (FMSI)    13
connectedness, lack of substitutes and concentration, lack of transparency, lever-
     age, market participant behaviour, information asymmetry and moral hazard. A
     vast number of papers are based on banking industry data, as it was considered the
     main source of systemic risk3. Since the beginning of the global financial crisis,
     many empirical studies have been performed on the basis of a more global ap-
     proach.

     There are several broad streams of studies that involve some kind of evaluation of
     systemic risk. There is a group of papers that, with the objective of measuring sys-
     temic risk, have developed Financial Stress Indexes (FSI) or fragility indexes. Some
     of these are coincident measures (like thermometers) that try to capture the level of
     financial stress on real time. Others are forward-looking indicators that, for example,
     calibrate the likelihood of simultaneous failure of a large number of financial inter-
     mediaries. The study of Illing and Liu (2006) can be considered as a seminal paper in
     this category. They develop a FSI for the Canadian financial system and propose
     several approaches to aggregate individual stress indicators into a composite stress
     index. Other relevant papers are Nelson and Perli (2007), Kritzman et al. (2010), Cal-
     darelli, Elekdag and Lall (2011), and Holló, Kremer and Lo Duca (2012). Holló, Kre-
     mer and Lo Duca (2012) perform a Composite Indicator of Systemic Stress (CISS) for
     the euro area, based on data of five segments of European financial markets (equity
     markets, bond markets, money markets, financial intermediaries and forex markets).
     They compute the Cumulative Distribution Function (CDF) of fifteen variables and
     take into account potential cross-correlations between market segments.

     Other approaches have in common the definition of systemic risk as an extreme loss
     on a portfolio of assets related to financial intermediaries’ balance sheets. This defi-
     nition of systemic risk focuses on the financial health of intermediaries, rather than
     on monetary and credit conditions. Examples of this methodology can be found in
     Segoviano and Goodhart (2009), Acharya et al. (2010), Adrian and Brunnermeier
     (2011), Huang, Zhou and Zhu (2011), Gray and Jobst (2011), Brownlees and Engle
     (2012) and Hovakimian, Kane and Laeven (2012).

     During the global financial and economic crisis, and especially in the context of the
     European sovereign debt crisis, many studies focused on the phenomenon of conta-
     gion. Relevant papers in this topic are Forbes and Rigobon (2001), Hyde, Bredin,
     and Nguyen (2007), Diebold and Yilmaz (2009) and Caporin et al. (2013). Some stud-
     ies show that correlations tend to increase during market crashes. As a consequence,
     the exposure to different countries’ equity markets offers less diversification in
     down markets than in up markets. This pattern has been shown to apply in other
     industries also4 (affecting the returns of global industries, individual stocks, hedge
     funds and international bond markets). The presence of sudden regime shifts, con-
     sidered by some authors as a symptom of systemic risk, has also been tested by
     many studies. In general there is a perception that every economy shows two types
     of regimes: regimes of GDP growth and low volatility and regimes characterized by
     GDP contraction and high volatility (usually in the context of high uncertainty). Sev-
     eral papers show the existence of sudden regime shifts not only in the context of

     3   See, for example, Rodríguez-Moreno & Peña (2013)
     4   See Ferreira and Gama (2010), Hong, Tu, and Zhou (2003) or Cappiello, Engle, and Sheppard (2006).

14   Comisión Nacional del Mercado de Valores
GDP but also in other economic or financial areas of interest like short-term interest
rates, inflation or market turbulence5.

This paper introduces a Spanish Financial Market Stress Indicator (FMSI), similar
to the “Composite Indicator of Systemic Stress” that Holló, Kremer and Lo Duca
(2012) proposed for the euro area as a whole6. This kind of indicator, which can be
included in the group of Financial Stress Indicators (FSI), represents a coincident
measure of systemic risk and tries to quantify and summarize the stress in the Span-
ish financial system in a single statistic. Of course, this kind of approach may have
some disadvantages due to the potential excessive simplification in the evaluation
of systemic risk. However, it offers some useful characteristics. Firstly, it allows the
real-time evaluation of financial stress in the whole financial system and the identi-
fication of past episodes of financial stress. Secondly, it can provide the basis infor-
mation for an early warning signal model that assesses when the system may be
nearing a high financial stress episode. It can also be used to test the impact of any
policy measure regarding financial stability.

One of the major strengths of the Spanish FMSI is related to its ample coverage of
the financial system. As we stated earlier, one of the main lessons drawn from the
financial crisis was the importance of the financial sector as a whole and not only
the banking sector as a potential source of generating and propagating systemic risk.
Taking these considerations into account, and following the approach of Holló, Kre-
mer and Lo Duca (2012), we have computed information on six financial market
segments that we consider crucial to evaluating any potential source of financial
stress: the money market, equity market, bond market, financial intermediaries, de-
rivatives market and forex market7 (see figure 1). Although this indicator represents
a considerable improvement with respect to previous indicators in terms of the
quantity and quality of data and of its coverage of the financial system, it is neces-
sary to highlight two limitations. Firstly, financial intermediaries’ information is
basically banking sector information and only some insurance companies’ data has
been included. As part of its supervisory duties, the CNMV receives data on other
relevant financial intermediaries such as mutual funds and investment services
firms but not at the desired frequency (daily). Secondly, the indicator does not in-
clude information on financial infrastructure due to a lack of data.

Our Spanish FMSI comprises 18 market-based financial stress variables equally
split into the six financial segments mentioned above. Stress in financial markets is
characterised by the increase in uncertainty, the asymmetry of information and the
rise in the risk aversion among investors (preference for safer and more liquid as-
sets). Our 18 stress variables that, in general, represent changes in volatility, credit
spreads, liquidity and loss of value in different instruments can be considered as
good indicators of these characteristics of stress in financial markets.

Under the methodology of Holló, Kremer and Lo Duca (2012), we compute the em-
pirical CDF of each variable and construct a separate financial stress sub-index for

5   See Smith (2002), Kumar and Okimoto (2007) and Kritzman and Li (2010).
6   The Bank of Spain publishes a simpler version of this indicator in its Financial Stability Report (FSR). See
    box 1.1 in the May-13 FSR for details.
7   This market also includes information on oil prices.

                                                               A Spanish Financial Market Stress Index (FMSI)      15
each of the six financial segments considered. In order to aggregate these sub-indi-
     ces into the global Spanish FMSI we also apply basic portfolio theory, which is one
     of the most important methodological innovations of our reference paper. This
     portfolio-theoretic aggregation takes into account the time-varying cross-correla-
     tions between our six sub-indices. As a consequence of this methodology, our Span-
     ish FMSI puts relatively more weight on high stress financial situations, due to the
     fact that stress tends to be high in several market segments at the same time. At the
     end, we try to capture the situation when financial instability is spread across the
     whole financial system after systemic risk materializes.

     Spanish financial system structure and supervision scheme                                                FIGURE 1

                                                       Blocks

           Financial intermediaries                    Markets                              Infrastructures

                   Banks                            Equity market
          Insurance/Pension funds                   Bond market                         Payment systems
                Mutual funds                       Money market                        Settlement system
          Investment services firms               Derivatives market                     Clearing system
                   Other                            Forex market

                Bank of Spain
                   CNMV
                                                       CNMV*                            CNMV/Bank of Spain
            Directorate Generale of
         Insurance and Pension Funds

                                                     Supervisor

     Source: CNMV. (*) The CNMV is not the supervisor of all submarkets presented in the Markets Block.

     This study also connects with the literature related to switching regimes that was
     presented earlier. We estimate an autoregressive Markov-switching model in order
     to identify the potential existence of different financial stress regimes according to
     the data provided by the FMSI. This kind of methodology allows us to evaluate in
     real time the possibility of being near a high financial stress episode and, if this is
     the case, the adoption of relevant policy measures to mitigate the risk.

     Finally, and taking into account that the propagation of a systemic risk should have
     some economic impact (according to all definitions of systemic risk), we estimate a
     threshold vector autoregression (TVAR) to assess the interaction between our Span-
     ish FMSI and some measures of economic activity. We try to identify one FMSI
     threshold at or above which financial stress is really high and may have a strong
     negative effect on the real economy. The relationship between financial system and
     real economy has also been explored by many studies8.

     8     Davig and Hakkio (2010), Hubrich and Tetlow (2011) and Hartmann et al. (2012).

16   Comisión Nacional del Mercado de Valores
3        Statistical design of the Spanish FMSI

3.1     Selection of markets and variables

In order to measure systemic risk across the Spanish financial system, we consider
the money market, bond market, equity market, financial intermediaries, foreign
exchange market and derivatives market as good representations of different seg-
ments of the financial system. Each of these segments will be presented as a sub-
index of the FMSI and will provide specific information for our composite indicator.

We include three variables in each of the market segments, so that the composite
indicator comprises 18 individual stress indicators, with the aim of measuring sys-
temic risk in real time. For that purpose, we use data which is available on a daily
or weekly basis. We basically include asset return volatilities, risk spreads and li-
quidity indicators to capture the main symptoms of financial stress. Long‑term
variables have been computed in order to cover as many financial stress episodes
and business cycles as possible. The three variables in each market segment should
provide complementary information to the indicator, although we expect a high
correlation between them during episodes of high financial stress (Holló, Kremer
and Lo Duca (2012)).

In what follows, a brief description of each market, their compounded variables9
and the data source is presented and organized by the representative market seg-
ment10.

–	Money market: this sub-index should reflect liquidity and counterparty risk in
   the inter-bank market (Heider, Hoerova and Holthausen (2010) or Acharya
   and Skeie (2011)). In general, money market variables capture some features
   like flight-to-quality and flight-to-liquidity effects, as well as the price impacts of
   adverse selection problems in banking during stress periods.

       •	Realised volatility of the three-month Euribor rate: realized volatility cal-
          culated as the weekly average of absolute daily rate changes, transformed
          by its recursive sample CDF. Data start 30 Dec. 1998. Source: Thomson
          Datastream. The volatility can reflect features like flight-to-quality, flight-
          to-liquidity and/or increasing asymmetric information; therefore a posi-
          tive relationship with systemic risk is highly expected.

9     In order to be consistent the same number of variables has been included in each market. Since the sub-
      indices are computed as simple averages, under the assumption of normally distributed variables, the
      inclusion of one additional variable in one particular market would reduce the variance of the average
      of the sub-index.
10    Summary statistics of the raw variables are provided in the annex.

                                                              A Spanish Financial Market Stress Index (FMSI)    17
•	Interest rate spread between three-month Euribor and three-month Span-
              ish Treasury Bills: weekly average of daily data, transformed by its recur-
              sive sample CDF. Data start 30 Dec. 1998 and 24 Mar. 1988 respectively.
              Source: Thomson Datastream. This variable represents a measure of li-
              quidity and counterparty risk, and shows the convenience premium on
              short-term Treasury paper.

           •	Three-month Libor-OIS spread: weekly average of daily difference be-
              tween three-month OIS and three-month Libor data, transformed by its
              recursive sample CDF. Data start 30 Dec. 1998 and 17 May. 1999 respec-
              tively. Source: Thomson Datastream. It is a measurement of liquidity and
              credit risk and also reflects the risk premium associated with lending to
              commercial banks. Therefore spread increases can be interpreted as a
              signal of high vulnerability in the financial system.

     –	Bond market: Movements in this market are related to sovereign risk and con-
        cerns about solvency and liquidity conditions in the corporate bond market.
        They can also be a consequence of an increase in the uncertainty or the risk
        aversion of investors. In addition sudden variations of the variables included
        in this market will have considerable impact not only on financial institutions
        but also on households.

           •	Realised volatility of the Spanish ten-year benchmark government bond
              index: weekly average of absolute daily yield changes, transformed by its
              recursive sample CDF. Data start 4 Apr. 1991. Source: Thomson Data-
              stream. Increases in the volatility can be a consequence of investor’s con-
              cerns about Government default risk.

           •	Yield spread between the Spanish ten-year government bond and Ger-
              man ten-year government bond: weekly average of daily difference be-
              tween Spanish and German ten-year bonds, transformed by its recursive
              sample CDF. Data start 4 Apr. 1991 and 1 Jan. 1980 respectively. Source:
              Thomson Datastream. This variable is a measure of sovereign risk pre-
              mium as long as the German bond is considered the safest and most liq-
              uid sovereign bond of the euro area.

           •	Bid-ask spread of Spanish government bonds: weekly average of daily
              bid-ask spread, transformed by its recursive sample CDF. Data start 11
              Aug. 1997. Source: Bloomberg. This variable reflects liquidity conditions
              in bond markets.

     –	Equity markets: equity market variables capture shifts in volatility, liquidity
        and sudden asset price movements that are common in periods of financial
        stress.

           •	Volatility of Spanish non-financial corporation index: weekly average of
              absolute daily log returns of the non-financial sector stock market index,
              transformed by its recursive sample CDF. Data start 2 Mar. 1987. Source:
              Thomson Datastream. In general, asset price volatility indicators point to
              stress in the stocks markets.

18   Comisión Nacional del Mercado de Valores
•	CMAX of Spanish non-financial corporation index: weekly average of
         daily maximum cumulated index losses of Spanish non-financial corpora-
         tion index, over a moving two-year window, transformed by its recursive
         sample CDF. Data start 2 Mar. 1987. Source: Thomson Datastream. The
         CMAX11 measurement is used to determine periods of crisis in interna-
         tional equity markets (Patel and Sarkar (1998) and Coudert and Gex
         (2006)), but most recently is often used as an input in stress indicators
         (Illing and Lui ,2006). Significant falls in price assets are captured by high
         levels of this variable.

      •	Ibex 35 liquidity: weekly average of daily bid-ask spread, transformed by
         its recursive sample CDF. Data start 9 Jul. 200312. Source: Thomson Data-
         stream. High financial stress levels are usually accompanied by drops in
         equity liquidity.

–	Financial intermediaries: Financial intermediaries play a major role in the
   correct functioning of the financial system. High increases in stress conditions
   for these institutions can be spread across the financial system and potentially
   have a strong negative impact on the real economy. The variables included in
   this market refer to volatility, credit risk and price movements.

      •	Realised volatility of the idiosyncratic equity return of the banking sector
         market index relative to Ibex 35 returns: idiosyncratic return calculated
         as the intercept from an OLS regression of daily log bank returns on the
         log market return over a moving two-year window; realized volatility cal-
         culated as the weekly average of absolute daily idiosyncratic returns,
         transformed by its recursive sample CDF. Data start 2 Mar. 1987. Source:
         Thomson Datastream. Increases in this indicator are interpreted as inves-
         tors’ experiencing high uncertainty and/or concerns about banking sector
         default risk.

      •	Financial sector credit risk spread: weekly average of daily CDS of five
         important Spanish banks, transformed by its recursive sample CDF. Data
         start 2 Jul. 200713. Source: Thomson Datastream. High values in risk pre-
         mium of these institutions imply a worsening in financing conditions
         that could be disseminated across the economy.

      •	CMAX of financial sector index combined with the inverse of its price-
         book ratio: weekly average of daily maximum cumulated index losses of
         Spanish financial sector index, over a moving two-year window and the
         inverse of the price-book ratio of these market, both transformed by their
         recursive sample CDF and then multiplied together. The final variable is
         obtained by taking the square root of this product. Data start 2 Mar. 1987
         and 1 Jan. 1990 respectively. Source: Thomson Datastream. High values

11                                              , where T=104 for weekly data.
12   From 1 Jan. 1999 to 9 Jul. 2003 Ibex 35 liquidity has been computed from Ibex 35’s constituents; weekly
     average of daily bid-ask spreads. Time varying weights have been used.
13   From 1 Jan. 1991 to 1 Jul. 2007 financial sector credit risk spread has been estimated from the yield
     spread between European A-rated financial corporations and the ten-year Spanish government bond.

                                                              A Spanish Financial Market Stress Index (FMSI)   19
of this variable are a consequence of high values in CMAX and in price-
                 book ratio, which means that the present market value of a corporation
                 has fallen significantly below its book value.

     –	Foreign Exchange market: this sub-index reflects large movements in foreign
        exchange markets. These movements are particularly relevant for those insti-
        tutions heavily dependent on non-domestic liabilities and also for those with a
        high exposure to non-domestic assets.

           •	Realised volatility of the euro exchange rate vis-à-vis the US dollar: weekly
              average of absolute daily log foreign exchange returns, transformed by its
              recursive sample CDF. Data start 1 Jan. 1980. Source: Thomson Datastream.

           •	Realised volatility of the euro exchange rate vis-à-vis the Japanese Yen:
              weekly average of absolute daily log foreign exchange returns, trans-
              formed by its recursive sample CDF. Data start 1 Jan. 1980. Source: Thom-
              son Datastream.

           •	Realised volatility of the euro exchange rate vis-à-vis the British Pound:
              weekly average of absolute daily log foreign exchange returns, trans-
              formed by its recursive sample CDF. Data start 1 Jan. 1980. Source: Thom-
              son Datastream.

     –	Derivatives market: Derivatives markets represent a special segment of the
        financial system in the sense that they are based upon another market, the
        underlying market. Their potential role in systemic risk was recognised by
        authorities during the last crisis, and has prompted some reforms, for example,
        in the OTC derivatives segment. The fluctuation of some relevant indicators of
        these markets can also be interpreted as signs of increasing uncertainty, risk
        aversion and financial stress.

           •	Realised volatility of IBEX-35 options: weekly average of daily implicit
              volatility of the IBEX 35 index, transformed by its recursive sample CDF.
              Data start 4 Jan. 1999. Source: Thomson Datastream.

           •	Realised volatility of IBEX-35 future open position: weekly average of
              daily volatility of the MEFF-IBEX 35 open interest index over a 60-day
              moving window, transformed by its recursive sample CDF. Data start 20
              Apr. 1992. Source: Thomson Datastream.

           •	Realised volatility of commodities index: weekly average of daily oil price
              volatility, transformed by its recursive sample CDF. Data start 1 Jan. 1980.
              Source: Thomson Datastream.

     3.2    Construction of market sub-indices

     In order to obtain a unique sub-index for each of the representative markets, it is
     necessary to transform each raw variable into a standardized one and then aggre-
     gate these new variables. The academic literature suggests several methodologies

20   Comisión Nacional del Mercado de Valores
for transforming the variables. For example, Morris (2010) proposes an empirical
normalization which consists of subtracting the sample mean of the variable and
dividing this difference by the sample standard deviation. Louzis and Vouldis
(2012) follow a logistic transformation in order to standardize the raw variables.
However, these approaches are based on the assumption of normally distributed
variables, an assumption that is often violated by the nature of the financial market
indicators. This fact implies some kind of robustness problems with the composite
indicator. Requiring robustness in the time dimension is an important issue in or-
der to create a real-time indicator, especially if it is going to be used as an early
warning signal. For this reason we have used the transformation based in the em-
pirical cumulative distribution function (CDF), such as in Holló, Kremer and Lo
Duca (2012).

Let us denote the data set of a particular variable xt as x = (x1,x2,…,xn) with n the total
number of observations in the sample. The ordered sample is denoted (x[1],x[2],…,x[n])
where x[1] ≤ x[2] ≤ … ≤ x[n] and [r] is referred to as the ranking number assigned to a
particular realisation of xt. The values in the original data set are arranged such that
x[n] represents the sample maximum and x[1] represents the sample minimum. The
transformed variables zt are then computed from the original variables xt on the
basis of the empirical CDF Fn(xt) as follows:

for t = 1,2,…,n. The empirical CDF Fn(x*) measures the total number of observations
xt not exceeding a particular value x* (which equals the corresponding ranking num-
ber x*) divided by the total number of observations in the sample (see Spanos
(1999)). If a value x occurs more than once, the ranking number assigned to each of
the observations is set to the average ranking. The empirical CDF is hence a function
which is non-decreasing and piecewise constant with jumps being multiples of 1/n
at the observed points. This results in transformed variables which are unit-free and
measured on an ordinal scale with range (0,1].

This transformation is applied recursively over expanding samples in order to fea-
ture the real-time character of the indicator. The pre-recursion period for each vari-
able runs from its first historical value to 4 January 2002, and all subsequent obser-
vations are transformed recursively on the basis of ordered samples recalculated
with one new observation added at a time:

for T = 1,2,…,N with N indicating the end of the full data sample.

Once the transformation of the three stress factors (j = 1,2,3) for each market
(i = 1,2,3,4,5,6) is computed, we end up with a data set of 18 homogenised stress

                                                    A Spanish Financial Market Stress Index (FMSI)   21
factors. In order to obtain markets’ sub-indices, we perform the arithmetic average14
     of the homogenized stress factors, which implies that each factor is equally weight-
     ed within the sub-index.

                                                              1 3
                                                              3∑
                                                   si , j =          zi , j , t
                                                                j =1

     3.3     Aggregation of sub-indices into the composite indicator

     The next step is the aggregation of the six sub-indices into a simple indicator to
     measure the systemic stress. Following standard portfolio theory; we have taken
     into account cross-correlations between individual assets returns. The methodology
     was proposed by Holló, Kremer and Lo Duca (2012) and also implemented by and
     Louzis and Vouldis, (2012), Milwood (2012) and Cabrera et al. (2014), for the design
     of systemic risk indicators in Greece, Jamaica and Colombia, respectively. Proceed-
     ing under this theory, the composite indicator puts more emphasis on situations
     where stress is predominant in several markets at the same time. The idea underly-
     ing this approach is to capture systemic risk in the sense that high financial instabil-
     ity is disseminated across the financial system.

     Each sub-index weight can be determined on the basis of the relative importance
     of this particular market for real economy activity15. The weights we have applied
     here are the following: 15% for money market, 20% for the bond market and eq-
     uity market, 30% for financial intermediaries, 5% for the foreign exchange market
     and 10% for derivatives. In order to incorporate the correlation between sub-indi-
     ces, we compute a unit-free indicator, bounded by the half-open interval (0, 1], ac-
     cording to:

                                       FMSI_ESPt = (w ° st) Ct (w ° st)'

     where w = (w1,w2,w3,w4,w5,w6) is defined as the vector of constant sub-index weights,
     st = (s1,t,s2,t,s3,t,s4,t,s5,t,s6,t) is the vector of sub-indices, and w ° st is the Hadamard-
     product16. Ct is the 6x6 matrix of time-varying cross-correlation coefficients ρij,t be-
     tween sub-indices i and j, represented as:

                                          1     ρ12,t    ρ13,t          ρ14,t     ρ15,t   ρ16,t
                                      ρ12,t       1      ρ23,t          ρ24,t     ρ25,t   ρ26,t
                                      ρ         ρ23,t         1         ρ34,t     ρ35,t   ρ36,t
                                 Ct = ρ13,t     ρ24,t    ρ34,t             1      ρ45,t   ρ46,t
                                       14,t
                                      ρ15,t     ρ25,t    ρ35,t          ρ45,t      1      ρ56,t
                                      ρ16,t     ρ26,t    ρ36,t          ρ46,t     ρ56,t    1

     14    The aggregation of the variables could be also done applying principal component analysis. It has been
           applied as a robustness test in section 3.5.
     15    The sub-index weight can be estimated from its relative average impact on industrial production growth
           calculated by a VAR model (see section 3.5).
     16    Element by element multiplication of the vector of sub-index weights and the vector of sub-index values
           in time t.

22   Comisión Nacional del Mercado de Valores
Time-varying cross-correlations ρij,t are performed recursively on the basis of expo-
nentially-weighted moving averages (EWMA) of respective covariances σij,t and
volatilities σi,t as approximated by the following formulas:

                                      σ ij ,t = λ σ ij ,t −1 + (1 − λ ) si ,t s j ,t

                                       σ i2,t = λ σ 2i ,t −1 + (1 − λ ) si ,t 2

                                             ρij ,t = σ ij ,t / σ i ,t σ j ,t

where i = 1,…6, j = 1,…6, t = 1,…T with si ,t = (si ,t − 0.5) represented the demeaned sub-
indices obtained by subtracting the theoretical mean of each indicator. The decay
factor or smoothing parameter λ is held constant through time at 0.93, while the
covariances and volatilities are initialized for t = 0 at their average values over the
pre-recursion period 1 January 1999 to 4 January 2002 (see figure 2).

Cross-correlations between sub-indices                                                                        FIGURE 2

                           ρ12               ρ13                   ρ14                   ρ15        ρ16
                           ρ23               ρ24                   ρ25                   ρ26        ρ34
                           ρ35               ρ36                   ρ45                   ρ46        ρ56
  1

 0.5

  0

-0.5

  -1
       Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-1 Jan-1 Jan- Jan- Jan- Jan-
        99   00   01   02   03   04   05   06   07   08   09    0     1    12   13   14   15

Source: Thomson Datastream, Bloomberg and CNMV. Cross-correlations are labelled as follows: 1 – money
market, 2 – bond market, 3 – equity market, 4 – financial intermediaries, 5 – foreign exchange markets, 6 – de-
rivatives. Weekly data from 1 Jan. 1999 to 6 Mar. 2015.

Periods in which all correlations are close to one (see 2009) can be considered as
extreme stress situations, as long as stress is spread across all financial markets.
Nevertheless, high values in pair correlations only indicate stress in two markets in
a certain period, which is not necessarily a signal of stress in the whole financial
system.

3.4      Backward extension

This section presents the Spanish FMSI obtained after the computation of the stress
variables and its aggregation according to the proposed methodology. Moreover, a
backward extension of the indicator is provided in order to verify if our FMSI is able
to capture important past events commonly known as high stress periods.

                                                                         A Spanish Financial Market Stress Index (FMSI)   23
The Spanish Financial Market Stress Indicator shown in figure 3 provides a char-
     acterization of systemic risk in one single number and also the contribution of
     each market segment to this risk. Remember that the simple aggregation of these
     contributions does not correspond to the composite indicator, because our refer-
     ence indicator takes into account cross market correlations. Both indicators tend
     to be similar when all our market segments are strongly correlated, usually in pe-
     riods of high stress. This was the case in 2008 with the Lehman Brothers collapse
     but, in general, our composite indicator will be lower than the sum of the contri-
     butions. According to the results presented in figure 3, we can see the first high
     value of the indicator (0.41) in Sep. 200117, after the terrorist attacks in the US.
     Equity markets and financial intermediaries experienced high stress, which was
     not observed in other segments of financial markets. The second episode of high
     stress took place at the end of 2008, when the indicator reached its historical
     maximum (0.87). Finally, in the context of the European sovereign debt crisis, fi-
     nancial markets suffered several periods of high financial stress. The stress was
     particularly high in the Spanish financial system by mid-2011 and mid-2012,
     when the indicator reached a value of 0.69. In these episodes, financial intermedi-
     aries, bond markets and equity markets were the most stressed segments. In addi-
     tion to the stress levels, figure 3 provides a first overview of correlation18 between
     markets.

     Financial Market Stress Indicator (FMSI)                                                         FIGURE 3

                    Money market                   Bond market                        Equity market
                    Financial intermediaries       Foreign exchange market            Derivatives market
                    Correlation                    FMSI
      0.90                                                                                                 0.90

      0.70                                                                                                 0.70

      0.50                                                                                                 0.50

      0.30                                                                                                 0.30

      0.10                                                                                                 0.10

     -0.10                                                                                                 -0.10

     -0.30                                                                                                 -0.30

     -0.50                                                                                                 -0.50
             Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-
              99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15

     Source: Thomson Datastream, Bloomberg and CNMV. Weekly data from 1 Jan. 1999 to 6 Mar. 2015.

     In the time period prior to 1999, there were several significant episodes of financial
     stress that deserve our attention because of its potential relevance in terms of sys-
     temic risk. We have performed a proxy for the FMSI which starts in Apr. 1987 with
     the historical data available in order to address this issue. The indicator presents
     some limitations due to the lack of data in some markets but in general it can be
     considered a good representation of financial stress in those years. As we can see
     from figure 4, where original and backward-extended FMSI are presented, at least

     17   These events have been studied in section 4.1.
     18   Correlation measured as the difference between the FMSI and a hypothetical perfect correlated FMSI.

24   Comisión Nacional del Mercado de Valores
two more periods of financial stress can be identified: one in Jan. 1991, when the
FMSI reached a value of 0.52, and the other one in Oct. 1992, with a value of 0.66.
These stress episodes are described in detail in section 4.1.

Backward-extended proxy-FMSI                                                                      FIGURE 4

                                            FMSI Proxy               FMSI
1.00

0.80

0.60

0.40

0.20

0.00
   Apr-87    Apr-90    Apr-93    Apr-96     Apr-99   Apr-02       Apr-05        Apr-08   Apr-11   Apr-14

Source: Thomson Datastream, Bloomberg and CNMV. Weekly data from 1 Apr. 1987 to 6 Mar. 2015.

Regarding correlation between sub-indices, we could say that during some periods
of stress several correlations have been near one (which implies perfect correla-
tion). In figure 5, the comparison between our reference FMSI and the indicator
which assumes perfect correlation shows that in moments of high financial stress
both indicators tend to be rather similar. On the contrary, in moments of low fi-
nancial stress, which in our sample could be associated with the period between
1997 and 2004, both indicators tend to diverge because of the low correlation be-
tween market segments. In general, it can be concluded that under low stress pe-
riods, market segments performance reflects the idiosyncratic characteristics of
each market.

FMSI versus hypothesis of perfect correlation                                                     FIGURE 5

                                          FMSI            Perfect correlation
1.00

0.80

0.60

0.40

0.20

0.00
   Apr-87    Apr-90    Apr-93    Apr-96     Apr-99       Apr-02    Apr-05       Apr-08   Apr-11    Apr-14

Source: Thomson Datastream, Bloomberg and CNMV. Weekly data from 1 Apr. 1987 to 6 Mar. 2015.

                                                             A Spanish Financial Market Stress Index (FMSI)   25
3.5    Robustness analysis

     The construction of any indicator of systemic risk implies the adoption of some subjec-
     tive decisions that can have significant consequences in terms of the properties of the
     indicator. We have performed some robustness checks in order to minimise some statis-
     tical problems. We are going to: (i) evaluate principal components analysis as aggrega-
     tion method of transformed variables, (ii) modify market weights using those estimated
     by vector autoregression (VAR) models, (iii) change the value of the smoothing param-
     eter and (iv) compare the results under recursive and non-recursive samples. In general,
     we conclude that our original Spanish FMSI is markedly robust and stable over time.

     –	Principal component analysis (PCA): PCA is a statistical technique to simplify
        a data set that was developed by Pearson (1901) and Hotelling (1933). This
        technique transforms a large number of variables into a smaller number of
        uncorrelated (orthogonal) factors, called principal components. Each compo-
        nent is a linear combination of the original data and ordered in such a way that
        the first component accounts for the largest possible variance. We have com-
        puted a separate PCA for the variables of each sub-index and used the first
        component information to estimate the aggregate sub-index instead of the sim-
        ple average of the three variables. As long as the composite indicator is ranged
        (0,1], the sub-indices have to be also ranged between 0 and 1. In order to esti-
        mate the new aggregate sub-indices capturing the maximum variance, the vec-
        tor modulus should be 1. These sub-indices are computed as follows:

                                           sit = ai1zi1,t + ai 2zi 2,t + ai 3zi 3,t

                                                  s.t. ai'ai = ∑aij2 = 1
                                                                    j

                                  sit* =   ∑a z
                                            j
                                                 2
                                                 j ij ,t   = a12zi1,t + a22zi 2,t + a32zi 3,t

     	where sit* are the market sub-indices with i = 1,2,3,4,5,6, aj represents the principal
       component coefficients of the first eigenvector with j = 1,2,3 for each of the varia-
       bles belonging to the six reference markets and zij,t the original variable data set.

     Sub-indices aggregation by principal component analysis                                                  FIGURE 6

                                                           FMSI PCA              FMSI
     1.00

     0.80

     0.60

     0.40

     0.20

     0.00
        Jan-99     Jan-01      Jan-03           Jan-05          Jan-07        Jan-09        Jan-11   Jan-13     Jan-15

     Source: Thomson Datastream, Bloomberg and CNMV. Weekly data from 1 Jan. 1999 to 6 Mar. 2015.

26   Comisión Nacional del Mercado de Valores
Figure 6 displays the FMSI based on PCA aggregation methodology and the
  original FMSI. According to the estimates, both indicators are very similar and
  only small differences appear in 2002. See Coudert and Gex (2006) and Louiz
  and Vouldis (2012) for further information related to the application of PCA in
  risk indicators.

–	Market weights: the selection of market weights can be done on the basis of
   VAR models and Impulse Response Functions (IRF) which are able to quantify
   the potential impact of financial shocks on real economy19.

	The new financial market weights with this approach are as follows: 13%
  for the money market, 3% for the bond market, 18% for the equity market,
  26% for financial intermediaries, 34% for the foreign exchange market and
  7% for derivatives. In general, these weights are not very different from that
  used in the original FMSI except for the bond and forex markets. Figure 7
  presents a comparison between the original FMSI and the new-weighted
  FMSI. In general, both indicators are very similar. Only during the period
  2000-2002 and in 2012 some differences are observed. During 2000-2002,
  the stress in forex markets at the beginning of the Monetary Union had a
  strong impact in the new-weighted FMSI because of the significant weight
  of this market in the index (34%). On the contrary, in 2012, the stress ob-
  served in bond markets had a smaller impact in the new FMSI. Due to the
  fact that both indicators are rather similar and that the original weights are
  perceived as more realistic20, we still prefer the original version of the indi-
  cator.

Market weighted FMSI                                                                                FIGURE 7

                                              Weighted FMSI                FMSI
1.00

0.80

0.60

0.40

0.20

0.00
   Jan-99        Jan-01       Jan-03      Jan-05      Jan-07      Jan-09          Jan-11   Jan-13     Jan-15

Source: Thomson Datastream, Bloomberg and CNMV. Weekly data from 1 Jan. 1999 to 6 Mar. 2015.

19     See Special Feature C: “Systemic Risk Methodologies” in Financial Stability Review, June 2011 (ECB).
20     Weights from VAR and IRF are based on industrial production data. For industrial sectors the high rele-
       vance of forex markets that these models obtain makes sense, but perhaps they may be underestimat-
       ing the potential relevance of other financial markets such as bond markets in other important eco-
       nomic sectors such as services or construction.

                                                               A Spanish Financial Market Stress Index (FMSI)    27
–	Changes in the smoothing parameter: exponential weighted moving averages
        are applied in order to decrease or eliminate the influence of random varia-
        tions. Roberts (1959) defined λ as the smoothing parameter which determines
        the rate at which old data enter into the calculation of EWMA statistics. A large
        value of λ gives more weight to recent data and less weight to old data (and the
        contrary for small values of λ).

     	In Holló, Kremer and Lo Duca (2012), the decay factor or smoothing parameter
       is held to be constant through time at a level of 0.93, which is an intermediate
       value. However, the authors computed the FMSI for other values of λ: 0.97
       (high value) and 0.89 (low value). We have also estimated our Spanish FMSI
       for these three lambda values.

     	Figure 8 illustrates small differences between the new indicators. It seems to
       be that the FMSI with a low smoothing parameter (λ=0.89) displays wide
       swings and spikes while the FMSI with a high parameter (λ=0.97) loses some
       power in the identification of some periods of stress (e.g. Sep. 2001). However,
       the differences are almost insignificant, so we can conclude that changes in the
       smoothing parameter do not imply any relevant alteration in the general be-
       haviour of the indicator.

     Changes in the smoothing parameter                                                             FIGURE 8

                                        FMSI 0.97       FMSI 0.93            FMSI 0.89

     1.00

     0.80

     0.60

     0.40

     0.20

     0.00
        Jan-99     Jan-01      Jan-03       Jan-05    Jan-07        Jan-09     Jan-11    Jan-13       Jan-15

     Source: Thomson Datastream, Bloomberg and CNMV. Weekly data from 1 Jan. 1999 to 6 Mar. 2015.

     –	Recursive versus non-recursive sample: we have computed the FMSI on a
        non-recursive basis. This implies the computation of CDF over the whole sam-
        ple period (from Apr. 1987). The original FMSI, based on recursive empirical
        CDF starting in January 2002, and the non-recursive FMSI are very similar (see
        figure 9). There is only a small difference 2003.

28   Comisión Nacional del Mercado de Valores
Backward-extended proxy-FMSI: recursive versus non-recursive                                       FIGURE 9

                                          Non-recursive              Recursive

1.00

0.80

0.60

0.40

0.20

0.00
   Apr-87    Apr-90    Apr-93    Apr-96      Apr-99       Apr-02    Apr-05       Apr-08   Apr-11    Apr-14

Source: Thomson Datastream, Bloomberg and CNMV. Weekly data from 1 Jan. 1999 to 6 Mar. 2015.

                                                              A Spanish Financial Market Stress Index (FMSI)   29
4       Evaluation of the Spanish FMSI

4.1    Ability to identify stress events

The first exercise we can do to evaluate the Spanish FMSI is related to its ability to
identify past stress episodes. In theory, the indicator should increase sizeably after
a systemic risk event and reach unusually high levels. Hence, a formal evaluation of
the indicator should take into account what a sizeable increase is and provide a
definition of this unusually high level. Any kind of evaluation may experience type
I errors, that is the failure to identify a high-stress event, and type II errors, which
consists in a false identification of a high-stress event. Some evaluation approaches
rely on “crisis defined by events” and others rely on “crisis defined by quantitative
thresholds”. It is not possible to be sure that either of these approaches is the best
option because both of them present problems. The “crisis defined by events” ap-
proach may miss some stress episodes that do not originate from a certain crisis
event. The “crisis defined by quantitative thresholds” approach may incur type II
errors (it is not necessarily true that when the FMSI is above a threshold, there is a
systemic risk). Our evaluation, similar to Holló, Kremer and Lo Duca (2012), is based
on the analysis of the peaks of the Spanish FMSI during a very long time period. We
test if these peaks can be associated with historical periods of stress or systemic
events in order to verify potential type II errors (a false report of stress).

Spanish FMSI and major financial stress episodes                                                 FIGURE 10

                                            FMSI Proxy               FMSI

1.00

0.80

0.60

0.40

0.20

0.00
    Apr-87   Apr-90     Apr-93    Apr-96    Apr-99       Apr-02   Apr-05     Apr-08     Apr-11     Apr-14

Source: CNMV. Weekly data from 1 Jan. 1999 to 6 Mar. 2015

Figure 10 illustrates the first significant hike of the historical Spanish FMSI in the
summer of 1990, coinciding with the invasion of Kuwait by Iraq. This conflict
prompted a sharp increase in oil prices and a decrease in risk appetite in global

                                                             A Spanish Financial Market Stress Index (FMSI)   31
markets. There followed a period of intermediate financial stress, related to this geo-
     political conflict. In 1992 there is a new historical maximum in the level of the indi-
     cator as a consequence of the European Exchange Rate Mechanism crisis. The huge
     tensions in the European exchange markets, which ended with the British Pound
     and the Italian Lira eventually leaving the system in September, were disseminated
     across global financial markets.

     In the second quarter of 1994, an unexpected change in the monetary policy of the
     Federal Reserve prompted a significant increase in long-term interest rates across
     the world. In the Spanish securities markets, the huge growth in sovereign bond
     yields changed the perception of investors. Investors, especially non-residents, sold
     a big proportion of their equity holdings and increase their investment in bonds.

     Between 1994 and 2001 there was a relatively long period of low stress in the finan-
     cial system which was interrupted in September 2001, after the terrorist attacks in
     the US. Afterwards, the indicator stayed in a mid-financial stress level; probably as
     a consequence of the accounting scandals of 2002 and 2003 (Enron and Worldcom
     were the most significant episodes).

     The maximum level of the Spanish FMSI was reached at the end of 2008, with the
     collapse of Lehman Brothers, although there was also a remarkable level of stress in
     financial markets in 2011-2012, during some episodes of turbulence in the context
     of the European sovereign debt crisis. The Spanish FMSI started to increase by mid-
     2007 when the first signs of the subprime crisis appeared and reached its historical
     maximum at the end of October (0.88), after the collapse of Lehman Brothers and
     the rescue of AIG. The high level of the indicator was maintained throughout the
     following weeks due to the uncertainty introduced by abandoning the plan to pur-
     chase toxic assets in the US. After that, the FMSI decreased sharply until it reached
     mid-levels of stress.

     The last period of stress showed by the indicator must be understood in the context
     of the European sovereign debt crisis, as was said before. This crisis was character-
     ised by the sharp increase of sovereign credit risk of those European countries per-
     ceived as more fragile in economic terms. During the crisis financial markets, and
     especially sovereign bond markets, equity markets and financial intermediaries suf-
     fered several episodes of extremely high stress. The first of these took place in May
     2010 and was related to the potential Greek default. The second one started in the
     spring of 2011, when the Portuguese government asked for financial assistance. In
     the summer of 2011, extreme volatility in the markets prompted the ban on short
     selling by various European securities regulators. The CNMV also adopted this
     measure for financial sector firms. The level of financial stress continued to be very
     high during the following months, due to the perception of a second recession in
     Europe, until the events that in 2012 drove the indicator to its second historical
     maximum since 1987. In June 2012, the Spanish Government solicited European
     financial assistance for its banking sector and in July the CNMV adopted a second
     ban on short selling, which was applied to financial and non-financial firms. Al-
     though the level of systemic stress in Spain was really high in the summer of 2012,
     it was mainly concentrated in the bond market, the equity market and the financial
     intermediaries sector (banking). Cross-correlations between financial segments in-
     cluded in the FMSI were significantly lower than in the stress period related to the
     Lehman Brothers collapse, when cross-correlations were very high.

32   Comisión Nacional del Mercado de Valores
You can also read